arch/unicore32/mm/fault.c

   1 /*
   2  * linux/arch/unicore32/mm/fault.c
   3  *
   4  * Code specific to PKUnity SoC and UniCore ISA
   5  *
   6  * Copyright (C) 2001-2010 GUAN Xue-tao
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12 #include <linux/module.h>
  13 #include <linux/signal.h>
  14 #include <linux/mm.h>
  15 #include <linux/hardirq.h>
  16 #include <linux/init.h>
  17 #include <linux/kprobes.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/page-flags.h>
  20 #include <linux/sched.h>
  21 #include <linux/io.h>
  22
  23 #include <asm/system.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/tlbflush.h>
  26
  27 /*
  28  * Fault status register encodings.  We steal bit 31 for our own purposes.
  29  */
  30 #define FSR_LNX_PF              (1 << 31)
  31
  32 static inline int fsr_fs(unsigned int fsr)
  33 {
  34         /* xyabcde will be abcde+xy */
  35         return (fsr & 31) + ((fsr & (3 << 5)) >> 5);
  36 }
  37
  38 /*
  39  * This is useful to dump out the page tables associated with
  40  * 'addr' in mm 'mm'.
  41  */
  42 void show_pte(struct mm_struct *mm, unsigned long addr)
  43 {
  44         pgd_t *pgd;
  45
  46         if (!mm)
  47                 mm = &init_mm;
  48
  49         printk(KERN_ALERT "pgd = %p\n", mm->pgd);
  50         pgd = pgd_offset(mm, addr);
  51         printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
  52
  53         do {
  54                 pmd_t *pmd;
  55                 pte_t *pte;
  56
  57                 if (pgd_none(*pgd))
  58                         break;
  59
  60                 if (pgd_bad(*pgd)) {
  61                         printk("(bad)");
  62                         break;
  63                 }
  64
  65                 pmd = pmd_offset((pud_t *) pgd, addr);
  66                 if (PTRS_PER_PMD != 1)
  67                         printk(", *pmd=%08lx", pmd_val(*pmd));
  68
  69                 if (pmd_none(*pmd))
  70                         break;
  71
  72                 if (pmd_bad(*pmd)) {
  73                         printk("(bad)");
  74                         break;
  75                 }
  76
  77                 /* We must not map this if we have highmem enabled */
  78                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
  79                         break;
  80
  81                 pte = pte_offset_map(pmd, addr);
  82                 printk(", *pte=%08lx", pte_val(*pte));
  83                 pte_unmap(pte);
  84         } while (0);
  85
  86         printk("\n");
  87 }
  88
  89 /*
  90  * Oops.  The kernel tried to access some page that wasn't present.
  91  */
  92 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
  93                 unsigned int fsr, struct pt_regs *regs)
  94 {
  95         /*
  96          * Are we prepared to handle this kernel fault?
  97          */
  98         if (fixup_exception(regs))
  99                 return;
 100
 101         /*
 102          * No handler, we'll have to terminate things with extreme prejudice.
 103          */
 104         bust_spinlocks(1);
 105         printk(KERN_ALERT
 106                "Unable to handle kernel %s at virtual address %08lx\n",
 107                (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 108                "paging request", addr);
 109
 110         show_pte(mm, addr);
 111         die("Oops", regs, fsr);
 112         bust_spinlocks(0);
 113         do_exit(SIGKILL);
 114 }
 115
 116 /*
 117  * Something tried to access memory that isn't in our memory map..
 118  * User mode accesses just cause a SIGSEGV
 119  */
 120 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
 121                 unsigned int fsr, unsigned int sig, int code,
 122                 struct pt_regs *regs)
 123 {
 124         struct siginfo si;
 125
 126         tsk->thread.address = addr;
 127         tsk->thread.error_code = fsr;
 128         tsk->thread.trap_no = 14;
 129         si.si_signo = sig;
 130         si.si_errno = 0;
 131         si.si_code = code;
 132         si.si_addr = (void __user *)addr;
 133         force_sig_info(sig, &si, tsk);
 134 }
 135
 136 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 137 {
 138         struct task_struct *tsk = current;
 139         struct mm_struct *mm = tsk->active_mm;
 140
 141         /*
 142          * If we are in kernel mode at this point, we
 143          * have no context to handle this fault with.
 144          */
 145         if (user_mode(regs))
 146                 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
 147         else
 148                 __do_kernel_fault(mm, addr, fsr, regs);
 149 }
 150
 151 #define VM_FAULT_BADMAP         0x010000
 152 #define VM_FAULT_BADACCESS      0x020000
 153
 154 /*
 155  * Check that the permissions on the VMA allow for the fault which occurred.
 156  * If we encountered a write fault, we must have write permission, otherwise
 157  * we allow any permission.
 158  */
 159 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
 160 {
 161         unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
 162
 163         if (!(fsr ^ 0x12))      /* write? */
 164                 mask = VM_WRITE;
 165         if (fsr & FSR_LNX_PF)
 166                 mask = VM_EXEC;
 167
 168         return vma->vm_flags & mask ? false : true;
 169 }
 170
 171 static int __do_pf(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 172                 struct task_struct *tsk)
 173 {
 174         struct vm_area_struct *vma;
 175         int fault;
 176
 177         vma = find_vma(mm, addr);
 178         fault = VM_FAULT_BADMAP;
 179         if (unlikely(!vma))
 180                 goto out;
 181         if (unlikely(vma->vm_start > addr))
 182                 goto check_stack;
 183
 184         /*
 185          * Ok, we have a good vm_area for this
 186          * memory access, so we can handle it.
 187          */
 188 good_area:
 189         if (access_error(fsr, vma)) {
 190                 fault = VM_FAULT_BADACCESS;
 191                 goto out;
 192         }
 193
 194         /*
 195          * If for any reason at all we couldn't handle the fault, make
 196          * sure we exit gracefully rather than endlessly redo the fault.
 197          */
 198         fault = handle_mm_fault(mm, vma, addr & PAGE_MASK,
 199                             (!(fsr ^ 0x12)) ? FAULT_FLAG_WRITE : 0);
 200         if (unlikely(fault & VM_FAULT_ERROR))
 201                 return fault;
 202         if (fault & VM_FAULT_MAJOR)
 203                 tsk->maj_flt++;
 204         else
 205                 tsk->min_flt++;
 206         return fault;
 207
 208 check_stack:
 209         if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
 210                 goto good_area;
 211 out:
 212         return fault;
 213 }
 214
 215 static int do_pf(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 216 {
 217         struct task_struct *tsk;
 218         struct mm_struct *mm;
 219         int fault, sig, code;
 220
 221         tsk = current;
 222         mm = tsk->mm;
 223
 224         /*
 225          * If we're in an interrupt or have no user
 226          * context, we must not take the fault..
 227          */
 228         if (in_atomic() || !mm)
 229                 goto no_context;
 230
 231         /*
 232          * As per x86, we may deadlock here.  However, since the kernel only
 233          * validly references user space from well defined areas of the code,
 234          * we can bug out early if this is from code which shouldn't.
 235          */
 236         if (!down_read_trylock(&mm->mmap_sem)) {
 237                 if (!user_mode(regs)
 238                     && !search_exception_tables(regs->UCreg_pc))
 239                         goto no_context;
 240                 down_read(&mm->mmap_sem);
 241         } else {
 242                 /*
 243                  * The above down_read_trylock() might have succeeded in
 244                  * which case, we'll have missed the might_sleep() from
 245                  * down_read()
 246                  */
 247                 might_sleep();
 248 #ifdef CONFIG_DEBUG_VM
 249                 if (!user_mode(regs) &&
 250                     !search_exception_tables(regs->UCreg_pc))
 251                         goto no_context;
 252 #endif
 253         }
 254
 255         fault = __do_pf(mm, addr, fsr, tsk);
 256         up_read(&mm->mmap_sem);
 257
 258         /*
 259          * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
 260          */
 261         if (likely(!(fault &
 262                (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 263                 return 0;
 264
 265         if (fault & VM_FAULT_OOM) {
 266                 /*
 267                  * We ran out of memory, call the OOM killer, and return to
 268                  * userspace (which will retry the fault, or kill us if we
 269                  * got oom-killed)
 270                  */
 271                 pagefault_out_of_memory();
 272                 return 0;
 273         }
 274
 275         /*
 276          * If we are in kernel mode at this point, we
 277          * have no context to handle this fault with.
 278          */
 279         if (!user_mode(regs))
 280                 goto no_context;
 281
 282         if (fault & VM_FAULT_SIGBUS) {
 283                 /*
 284                  * We had some memory, but were unable to
 285                  * successfully fix up this page fault.
 286                  */
 287                 sig = SIGBUS;
 288                 code = BUS_ADRERR;
 289         } else {
 290                 /*
 291                  * Something tried to access memory that
 292                  * isn't in our memory map..
 293                  */
 294                 sig = SIGSEGV;
 295                 code = fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR;
 296         }
 297
 298         __do_user_fault(tsk, addr, fsr, sig, code, regs);
 299         return 0;
 300
 301 no_context:
 302         __do_kernel_fault(mm, addr, fsr, regs);
 303         return 0;
 304 }
 305
 306 /*
 307  * First Level Translation Fault Handler
 308  *
 309  * We enter here because the first level page table doesn't contain
 310  * a valid entry for the address.
 311  *
 312  * If the address is in kernel space (>= TASK_SIZE), then we are
 313  * probably faulting in the vmalloc() area.
 314  *
 315  * If the init_task's first level page tables contains the relevant
 316  * entry, we copy the it to this task.  If not, we send the process
 317  * a signal, fixup the exception, or oops the kernel.
 318  *
 319  * NOTE! We MUST NOT take any locks for this case. We may be in an
 320  * interrupt or a critical region, and should only copy the information
 321  * from the master page table, nothing more.
 322  */
 323 static int do_ifault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 324 {
 325         unsigned int index;
 326         pgd_t *pgd, *pgd_k;
 327         pmd_t *pmd, *pmd_k;
 328
 329         if (addr < TASK_SIZE)
 330                 return do_pf(addr, fsr, regs);
 331
 332         if (user_mode(regs))
 333                 goto bad_area;
 334
 335         index = pgd_index(addr);
 336
 337         pgd = cpu_get_pgd() + index;
 338         pgd_k = init_mm.pgd + index;
 339
 340         if (pgd_none(*pgd_k))
 341                 goto bad_area;
 342
 343         pmd_k = pmd_offset((pud_t *) pgd_k, addr);
 344         pmd = pmd_offset((pud_t *) pgd, addr);
 345
 346         if (pmd_none(*pmd_k))
 347                 goto bad_area;
 348
 349         set_pmd(pmd, *pmd_k);
 350         flush_pmd_entry(pmd);
 351         return 0;
 352
 353 bad_area:
 354         do_bad_area(addr, fsr, regs);
 355         return 0;
 356 }
 357
 358 /*
 359  * This abort handler always returns "fault".
 360  */
 361 static int do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 362 {
 363         return 1;
 364 }
 365
 366 static int do_good(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 367 {
 368         unsigned int res1, res2;
 369
 370         printk("dabt exception but no error!\n");
 371
 372         __asm__ __volatile__(
 373                         "mff %0,f0\n"
 374                         "mff %1,f1\n"
 375                         : "=r"(res1), "=r"(res2)
 376                         :
 377                         : "memory");
 378
 379         printk(KERN_EMERG "r0 :%08x  r1 :%08x\n", res1, res2);
 380         panic("shut up\n");
 381         return 0;
 382 }
 383
 384 static struct fsr_info {
 385         int (*fn) (unsigned long addr, unsigned int fsr, struct pt_regs *regs);
 386         int sig;
 387         int code;
 388         const char *name;
 389 } fsr_info[] = {
 390         /*
 391          * The following are the standard Unicore-I and UniCore-II aborts.
 392          */
 393         { do_good,      SIGBUS,  0,             "no error"              },
 394         { do_bad,       SIGBUS,  BUS_ADRALN,    "alignment exception"   },
 395         { do_bad,       SIGBUS,  BUS_OBJERR,    "external exception"    },
 396         { do_bad,       SIGBUS,  0,             "burst operation"       },
 397         { do_bad,       SIGBUS,  0,             "unknown 00100"         },
 398         { do_ifault,    SIGSEGV, SEGV_MAPERR,   "2nd level pt non-exist"},
 399         { do_bad,       SIGBUS,  0,             "2nd lvl large pt non-exist" },
 400         { do_bad,       SIGBUS,  0,             "invalid pte"           },
 401         { do_pf,        SIGSEGV, SEGV_MAPERR,   "page miss"             },
 402         { do_bad,       SIGBUS,  0,             "middle page miss"      },
 403         { do_bad,       SIGBUS,  0,             "large page miss"       },
 404         { do_pf,        SIGSEGV, SEGV_MAPERR,   "super page (section) miss" },
 405         { do_bad,       SIGBUS,  0,             "unknown 01100"         },
 406         { do_bad,       SIGBUS,  0,             "unknown 01101"         },
 407         { do_bad,       SIGBUS,  0,             "unknown 01110"         },
 408         { do_bad,       SIGBUS,  0,             "unknown 01111"         },
 409         { do_bad,       SIGBUS,  0,             "addr: up 3G or IO"     },
 410         { do_pf,        SIGSEGV, SEGV_ACCERR,   "read unreadable addr"  },
 411         { do_pf,        SIGSEGV, SEGV_ACCERR,   "write unwriteable addr"},
 412         { do_pf,        SIGSEGV, SEGV_ACCERR,   "exec unexecutable addr"},
 413         { do_bad,       SIGBUS,  0,             "unknown 10100"         },
 414         { do_bad,       SIGBUS,  0,             "unknown 10101"         },
 415         { do_bad,       SIGBUS,  0,             "unknown 10110"         },
 416         { do_bad,       SIGBUS,  0,             "unknown 10111"         },
 417         { do_bad,       SIGBUS,  0,             "unknown 11000"         },
 418         { do_bad,       SIGBUS,  0,             "unknown 11001"         },
 419         { do_bad,       SIGBUS,  0,             "unknown 11010"         },
 420         { do_bad,       SIGBUS,  0,             "unknown 11011"         },
 421         { do_bad,       SIGBUS,  0,             "unknown 11100"         },
 422         { do_bad,       SIGBUS,  0,             "unknown 11101"         },
 423         { do_bad,       SIGBUS,  0,             "unknown 11110"         },
 424         { do_bad,       SIGBUS,  0,             "unknown 11111"         }
 425 };
 426
 427 void __init hook_fault_code(int nr,
 428                 int (*fn) (unsigned long, unsigned int, struct pt_regs *),
 429                 int sig, int code, const char *name)
 430 {
 431         if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
 432                 BUG();
 433
 434         fsr_info[nr].fn   = fn;
 435         fsr_info[nr].sig  = sig;
 436         fsr_info[nr].code = code;
 437         fsr_info[nr].name = name;
 438 }
 439
 440 /*
 441  * Dispatch a data abort to the relevant handler.
 442  */
 443 asmlinkage void do_DataAbort(unsigned long addr, unsigned int fsr,
 444                         struct pt_regs *regs)
 445 {
 446         const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
 447         struct siginfo info;
 448
 449         if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
 450                 return;
 451
 452         printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 453                inf->name, fsr, addr);
 454
 455         info.si_signo = inf->sig;
 456         info.si_errno = 0;
 457         info.si_code = inf->code;
 458         info.si_addr = (void __user *)addr;
 459         uc32_notify_die("", regs, &info, fsr, 0);
 460 }
 461
 462 asmlinkage void do_PrefetchAbort(unsigned long addr,
 463                         unsigned int ifsr, struct pt_regs *regs)
 464 {
 465         const struct fsr_info *inf = fsr_info + fsr_fs(ifsr);
 466         struct siginfo info;
 467
 468         if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
 469                 return;
 470
 471         printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
 472                inf->name, ifsr, addr);
 473
 474         info.si_signo = inf->sig;
 475         info.si_errno = 0;
 476         info.si_code = inf->code;
 477         info.si_addr = (void __user *)addr;
 478         uc32_notify_die("", regs, &info, ifsr, 0);
 479 }